CN107784320A - Radar range profile's target identification method based on convolution SVMs - Google Patents

Radar range profile's target identification method based on convolution SVMs Download PDF

Info

Publication number
CN107784320A
CN107784320A CN201710889757.4A CN201710889757A CN107784320A CN 107784320 A CN107784320 A CN 107784320A CN 201710889757 A CN201710889757 A CN 201710889757A CN 107784320 A CN107784320 A CN 107784320A
Authority
CN
China
Prior art keywords
svms
cnn
range profile
dimensional
convolution
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201710889757.4A
Other languages
Chinese (zh)
Other versions
CN107784320B (en
Inventor
沈晓峰
何旭东
廖阔
司进修
王莎
邓贝贝
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN201710889757.4A priority Critical patent/CN107784320B/en
Publication of CN107784320A publication Critical patent/CN107784320A/en
Application granted granted Critical
Publication of CN107784320B publication Critical patent/CN107784320B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G01S7/418—Theoretical aspects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Abstract

The invention belongs to Radar Technology field, more particularly to a kind of radar one-dimensional distance target identification method based on concatenated convolutional neutral net and SVMs, concrete scheme to be：First, random distance disturbance, extension and plus noise is carried out to the one-dimensional range profile that radar target back scattering simulation software obtains to handle, and as primitive character；Secondly, one hot codings are done to the label of sample data；Then, using the method for deep learning, concatenated convolutional SVMs is built in conjunction with SVMs using concatenated convolutional neutral net, high order parameters are extracted come training network C_CNN using stochastic gradient descent method；Then, small parameter perturbations are carried out to concatenated convolutional SVMs CCNN_SVM using tape label sample data；Finally, parameter model is obtained using the concatenated convolutional SVMs network training, and classification is identified to the sample identified.The correct recognition rata of the inventive method convolutional neural networks reaches 92.34%, and the correct recognition rata of depth convolution SVMs reaches 95.59%.

Description

Radar range profile's target identification method based on convolution SVMs
Technical field
The invention belongs to Radar Technology field, more particularly to it is a kind of based on concatenated convolutional neutral net and SVMs Radar one-dimensional distance target identification method.
Background technology
Differentiate it being one of effective way that distant object identifies to target classification by radar return data.
Radar range profile's reflect distribution of the target scattering center on radar line of sight, embody the shape knot of target The physical messages such as structure, and easily obtained using high resolution radar, therefore it is widely used in radar target recognition field.In recent years Come, good effect is obtained in field of image recognition based on the recognition methods of deep learning.It is different from and traditional is manually set spy The mode of sign, it can be learnt automatically from one-dimensional range profile using convolutional neural networks model special to the target high-order for being advantageous to classification Sign, classification learning is done to high-order feature using SVMs again on this basis, therefore, study based on concatenated convolutional support to Amount machine it is one-dimensional as target identification method is expected to further improve object recognition rate and improves the stability of model.
The content of the invention
The defects of for prior art, the invention provides a kind of new mesh based on concatenated convolutional supporting vector machine model Mark recognition methods.
The technical scheme is that：
First, random distance disturbance, extension are carried out to the one-dimensional range profile that radar target back scattering simulation software obtains And plus noise processing, and as primitive character；Secondly, one-hot codings are done to the label of sample data；Then, utilize The method of deep learning, concatenated convolutional SVMs is built in conjunction with SVMs using concatenated convolutional neutral net, High order parameters are extracted come training network C_CNN using stochastic gradient descent method；Then, cascade is rolled up using tape label sample data Product SVMs CCNN_SVM carries out small parameter perturbations；Finally, joined using the concatenated convolutional SVMs network training Exponential model, and classification is identified to the sample identified.
A kind of Radar range profile's target identification method based on convolution SVMs, its step are specific as follows：
S1, data set is obtained, be specially：
One-dimensional range profile data are obtained using high-resolution radar, the data set format of the one-dimensional range profile data isOrderThe tally set corresponding to one-dimensional range profile data set is represented, wherein, K tables Show target classification sum, F represents that the one-dimensional range profile feature of single target is counted out, NiThe i-th class target sample number is represented,For total sample number in data acquisition system,Represent the i-th classification target n-th it is one-dimensional away from From as having the sample of 320 Range Profile characteristic points, yin=[yin(1),yin(2),...,yin(K) one-hot coding staffs] are used Formula, i=1,2,3 ..., K, j represent sample class sum；
S2, pretreatment, division data set is training set and test set, specific as follows：
S21, to data set D in S1(0)Front and back end is carried out with carrying out random phase disturbance after the machine transplanting of rice 0, while by each sample 10 times of extensions of this progress, data set now is designated asWherein,
S22, to D described in S21(1)Plus noise processing is carried out, energy normalized is then carried out, by the sample after normalization This collection is designated as
S23, by D described in S22(2)In similar target sample according to 7:3 ratio random division composing training collection and test Collection, note training set areRemember that test set isWherein,I-th classification target the n-th width one-dimensional range profile sample is represented, and dimension is 400, BiRepresent test The i-th classification target one-dimensional range profile number is concentrated,For total sample number in test set, andFor number According to collection total sample number；
S3, make shape remodeling to sample data, obtain being suitable for the Radar range profile's data of the shape of convolution, That is, shape remodeling is made respectively to ready-portioned data set in S2, the shape N*400 of one-dimensional radar range profile data is changed into It is suitable for doing the shape N*1*400*1 of convolution form；
S4, the one-dimensional concatenated convolutional neutral net C_CNN of structure, the input of the C_CNN is that the radar obtained in S3 is one-dimensional Range Profile data, the convolution kernel size of all convolutional layers is 1 × 3, and the core size of all pond layers is 1 × 11, C_CNN's Convolutional layer and full articulamentum activation primitive use linear correction function (ReLU) function, all convolution kernel weights initialisation modes Using Gauss normal distribution, and l2 regularizations are used, the characteristic vector of pond layer is input to full articulamentum, full articulamentum swashs Function living uses ReLU functions：
Pond method is set, it is contemplated that multiple peak regions be present in the one-dimensional range profile of radar target, therefore using maximum The mode in pond retains effective high communication number,
Level deep feature extraction network C NN_1 is built, wherein, the CNN_1 includes 3 convolution pond layers and 2 layers are complete Articulamentum, the output of last pond layer of the CNN_1 are the inputs of first full articulamentum, and the CNN_1's is last One pond layer step-length is 2, and remaining pond layer step-length is 1, and the full articulamentum nodes of the CNN_1 are set to 512,
Building two level depth characteristic extraction network C NN_2, the CNN_2 includes 3 convolution pond layers and 3 layers of full connection Layer, last pond layer of the CNN_2 are connected with first full articulamentum, last pond layer of the CNN_2 Step-length is 2, and remaining pond layer step-length is 1, and 2 nodes of the CNN_2 are set to 512 and 128 successively,
The output of last pond layer of the CNN_1 is the input of the CNN_2；
S5, structuring one-dimensional convolutional neural networks grader：Last full articulamentum connects softmax layers, i.e. grader letter Number uses softmax functions；
S6, the one-dimensional concatenated convolutional neutral net built according to S4 import training data, right respectively using gradient descent method CNN_1 and CNN_2 hyper parameter is finely adjusted, and after iteration S steps, obtains extracting the parameter model of feature, wherein, 100≤S≤ 200；
One S7, structure concatenated convolutional SVMs, it is specially：
The one-dimensional concatenated convolutional neutral net C_CNN built using S4, by the 2 of level deep feature extraction network C NN_1 Individual full articulamentum and two level depth characteristic extraction network C the NN_2 full articulamentum of last layer remove, and F_CNN are designated as, by F_ The feature input SVMs of CNN extractions, then high-order feature is learnt by the SVMs, while will be corresponding One-hot labels input SVMs after being transformed to ten's digit, form concatenated convolutional SVMs, are designated as CCNN_SVM；
S8, the concatenated convolutional SVMs parameter built using trellis search method to S7 are finely adjusted, training S steps Afterwards, concatenated convolutional SVMs network model to the end is obtained：
The parameter combination F_CNN networks of the depth characteristic extraction network obtained according to S6 carry out effective special to identification target Sign extraction, and the feature of extraction is input to SVMs,
Meanwhile the kernel function of SVMs uses RBF, hyperparameter optimization uses grid data service, to parameter C and gamma carries out optimizing in specified parameter area, wherein, parameter C represents the penalty coefficient in SVMs, gamma Represent the Gaussian kernel coefficient of RBF, parameter C Search Range be [1,3,10], gamma Search Range be (0.1, 0.3,0.5)；
S9, using the concatenated convolutional SVMs network model obtained in S8 to input sample carry out target identification.
Further, D is obtained described in S21(1)Concretely comprise the following steps：
Random sequence is produced by function randperm (Q), wherein, the length of the random sequence is Q=80；
To data set in S1In each width one-dimensional range profile, the side by front and back end with the machine transplanting of rice 0 Method carries out one-dimensional range profile phase perturbation, each width one-dimensional range profile is expanded into 10 width, the data set after note processing isWherein, it is described to data set D(0)Front and back end is carried out with the machine transplanting of rice 0, inserts 80 0 altogether.
Further, the plus noise processing described in S22, that is, add 22db white Gaussian noises, noise calculation formula can be with table It is shown asWherein, 400 be one-dimensional range profile dimension.
Further, specific method is finely tuned described in S6 is：
S61, the phenomenon for setting CNN_1 loss function loss1 auxiliary C_CNN to avoid gradient passback from disappearing；
S62, CNN_1 and CNN_2 loss function use this special loss function of logic, and its expression formula is：Wherein, n is input data quantity, ymFor corresponding sample Label, pmRepresent the probable value that model is calculated；
S63, the gradient of Adam methods is used to decline with adaptive different learning rates.
Further, S=200 described in S6 and S8.
The beneficial effects of the invention are as follows：
Original signal characteristic of the inventive method based on radar target-range image, it is carried out random phase disturbance, Sample extends and plus noise processing, afterwards as the robust of the input feature vector of identifying system, thus enhancing identifying system Property；Construct concatenated convolutional SVMs, combined on the basis of convolutional neural networks automatically extract validity feature support to Amount machine, improves discrimination.And identification test has been carried out to 5 classes emulation Aircraft Targets one-dimensional range profile data with this model, its The correct recognition rata of middle convolutional neural networks reaches 92.34%, and the correct recognition rata of depth convolution SVMs reaches 95.59%.
Brief description of the drawings
Fig. 1 is concatenated convolutional Artificial Neural Network Structures schematic diagram.
Fig. 2 is based on concatenated convolutional SVMs network architecture.
Fig. 3 concatenated convolutional SVMs identification process figures.
Embodiment
The present invention will be described below in conjunction with the accompanying drawings.
Using radar target back scattering simulation software to the class Aircraft Targets of An-26, B-1B, B-52, F-15, Tu-16 etc. 5 To be sampled at intervals of 0.1 degree of attitude angle, then each target produces 1800 width Range Profiles, produces totally 9000 width one-dimensional range profile number According to the dimension of every width Range Profile is 320 dimensions, and note raw data set is：Wherein, the i-th classification target n-th Width Range Profile is expressed as：
To every width one-dimensional range profile by front and back end with the machine transplanting of rice 0 (wherein insert 0 sum be 80), carry out phase perturbation, so as to Each width Range Profile is expanded into 10 width pictures, plus noise processing then is carried out using the formula in S22 to the data set after extension, 22db white Gaussian noises are added, and energy normalized processing is carried out to data set, the data acquisition system after note processing is：Wherein,
By D(2)In similar target sample according to 7:3 ratio random division composing training collection and test set, remember training set For：Wherein,Represent i-th classification target the n-th width dimension For 400 one-dimensional range profile sample；
WithRepresent sampleClass label vector, training sample tag set is designated asSimilarly note test set is：WithRepresent to survey Sample sheetClass label vector, test sample tag set is designated as
1. the model training stage
The concatenated convolutional neutral net as constructed by Fig. 1 is built using tensorflow, by training setAs input, shape remodeling is (63000, Isosorbide-5-Nitrae 00,1), inputs C_CNN networks, is utilized This special loss function of S62 logic optimizes to network performance, using the gradient descent method of adam methods to the C_CNN networks Parameter learnt, iteration S=200 step after, obtain network paramter models to the end.Utilize model parameter combination F_CNN Feature extraction is carried out to training data, feature is designated as F, instructed the feature extracted as the input of SVMs again Practice, optimizing is carried out to SVMs hyper parameter using grid data service, wherein, parameter C Search Range is [1,3,10], Gamma Search Range is that (0.1,0.3,0.5) obtains concatenated convolutional SVMs network model parameter.
2. target sample cognitive phase
By test set sampleIn each sample as the defeated of CCNN_SVM models Enter, then carry out target identification using the model, and obtain corresponding test set and differentiate probability vector output collectionForward prediction is carried out to the sample data of each input, obtains the output layer of single sample Vector is r=[r1,r2,...,r5]T, then the classification number for predicting sample to be identified isJ represents sample class Classification corresponding to the serial number of maximum in 5 neuron output values of sum, i.e. output layer.
Using the one-dimensional as target identification method based on concatenated convolutional SVMs of the emulation data verification present invention.It is imitative One-dimensional range profile data of the 5 class Aircraft Targets under different attitude angles really are generated, emulation data are pressed 7：3 ratio is random It has been divided into training dataset and test data set.By the way that each target in test set is identified, the present invention is obtained to 5 class targets Average correct recognition rata be about 95.59%.

Claims (5)

1. a kind of Radar range profile's target identification method based on convolution SVMs, it is characterised in that its step has Body is as follows：
S1, data set is obtained, be specially：
One-dimensional range profile data are obtained using high-resolution radar, the data set format of the one-dimensional range profile data isOrderThe tally set corresponding to one-dimensional range profile data set is represented, wherein, K tables Show target classification sum, F represents that the one-dimensional range profile feature of single target is counted out, NiThe i-th class target sample number is represented,For total sample number in data acquisition system,Represent the i-th classification target n-th it is one-dimensional away from From as having the sample of 320 Range Profile characteristic points, yin=[yin(1),yin(2),...,yin(K) one-hot coding staffs] are used Formula, i=1,2,3 ..., K, j represent sample class sum；
S2, pretreatment, division data set is training set and test set, specific as follows：
S21, to data set D in S1(0)Front and back end is carried out with progress random phase disturbance after the machine transplanting of rice 0, while each sample is entered 10 times of extensions of row, data set now is designated asWherein,
S22, to D described in S21(1)Plus noise processing is carried out, energy normalized is then carried out, by the sample set after normalization It is designated as
S23, by D described in S22(2)In similar target sample according to 7:3 ratio random division composing training collection and test set, note Training set isRemember that test set isWherein,I-th classification target the n-th width one-dimensional range profile sample is represented, and dimension is 400, BiRepresent test The i-th classification target one-dimensional range profile number is concentrated,For total sample number in test set, andFor number According to collection total sample number；
S3, make shape remodeling to sample data, obtain being suitable for the Radar range profile's data of the shape of convolution, i.e. right Ready-portioned data set makees shape remodeling respectively in S2, by the shape N*400 of one-dimensional radar range profile data, changes into suitable use In the form for the shape N*1*400*1 for doing convolution；
S4, the one-dimensional concatenated convolutional neutral net C_CNN of structure, the input of the C_CNN is the radar one-dimensional distance obtained in S3 As data, the convolution kernel size of all convolutional layers is 1 × 3, and the core size of all pond layers is 1 × 11, C_CNN convolution Layer and full articulamentum activation primitive use linear correction function (ReLU) function, and all convolution kernel weights initialisation modes use Gauss normal distribution, and l2 regularizations are used, the characteristic vector of pond layer is input to full articulamentum, the activation letter of full articulamentum Number uses ReLU functions：
Pond method is set, effective high communication number is retained by the way of maximum pond,
Level deep feature extraction network C NN_1 is built, wherein, the CNN_1 includes 3 convolution pond layers and 2 layers of full connection Layer, the output of last pond layer of the CNN_1 are the inputs of first full articulamentum, the CNN_1 last Pond layer step-length is 2, and remaining pond layer step-length is 1, and the full articulamentum nodes of the CNN_1 are set to 512,
The depth characteristic extraction of structure two level network C NN_2, the CNN_2 include 3 convolution pond layers and 3 layers of full articulamentum, institute Last the pond layer for stating CNN_2 is connected with first full articulamentum, and last pond layer step-length of the CNN_2 is 2, remaining pond layer step-length is 1, and 2 nodes of the CNN_2 are set to 512 and 128 successively,
The output of last pond layer of the CNN_1 is the input of the CNN_2；
S5, structuring one-dimensional convolutional neural networks grader：Last full articulamentum connects softmax layers, i.e. classifier functions make With softmax functions；
S6, the one-dimensional concatenated convolutional neutral net built according to S4 import training data, using gradient descent method respectively to CNN_1 It is finely adjusted with CNN_2 hyper parameter, after iteration S steps, obtains extracting the parameter model of feature, wherein, 100≤S≤200；
One S7, structure concatenated convolutional SVMs, it is specially：
The one-dimensional concatenated convolutional neutral net C_CNN built using S4 is complete by 2 of level deep feature extraction network C NN_1 Articulamentum and two level depth characteristic extraction network C the NN_2 full articulamentum of last layer remove, and are designated as F_CNN, F_CNN is carried The feature input SVMs taken, then high-order feature is learnt by the SVMs, while will corresponding one-hot Label inputs SVMs after being transformed to ten's digit, forms concatenated convolutional SVMs, is designated as CCNN_SVM；
S8, the concatenated convolutional SVMs parameter built using trellis search method to S7 are finely adjusted, and after training S steps, are obtained Concatenated convolutional SVMs network model to the end：
The parameter combination F_CNN networks of the depth characteristic extraction network obtained according to S6 carry out effective feature to identification target and carried Take, and the feature of extraction be input to SVMs,
Meanwhile the kernel function of SVMs uses RBF, hyperparameter optimization uses grid data service, to parameter C and Gamma carries out optimizing in specified parameter area, wherein, parameter C represents the penalty coefficient in SVMs, gamma tables Show the Gaussian kernel coefficient of RBF, parameter C Search Range is [1,3,10], gamma Search Range for (0.1,0.3, 0.5)；
S9, using the concatenated convolutional SVMs network model obtained in S8 to input sample carry out target identification.
2. a kind of Radar range profile's target identification method based on convolution SVMs according to claim 1, It is characterized in that：D is obtained described in S21(1)Concretely comprise the following steps：
Random sequence is produced by function randperm (Q), wherein, the length of the random sequence is Q=80；
To data set in S1In each width one-dimensional range profile, by front and back end with the machine transplanting of rice 0 method carry out One-dimensional range profile phase perturbation, each width one-dimensional range profile is expanded into 10 width, the data set after note processing isWherein, it is described to data set D(0)Front and back end is carried out with the machine transplanting of rice 0, inserts 80 0 altogether.
3. a kind of Radar range profile's target identification method based on convolution SVMs according to claim 1, It is characterized in that：Plus noise processing described in S22, that is, add 22db white Gaussian noises, noise calculation formula can be expressed asWherein, 400 be one-dimensional range profile dimension.
4. a kind of Radar range profile's target identification method based on convolution SVMs according to claim 1, It is characterized in that：Specific method is finely tuned described in S6 is：
S61, the phenomenon for setting CNN_1 loss function loss1 auxiliary C_CNN to avoid gradient passback from disappearing；
S62, CNN_1 and CNN_2 loss function use this special loss function of logic, and its expression formula is：Wherein, n is input data quantity, ymFor corresponding sample Label, pmRepresent the probable value that model is calculated；
S63, the gradient of Adam methods is used to decline with adaptive different learning rates.
5. a kind of Radar range profile's target identification method based on convolution SVMs according to claim 1, It is characterized in that：S=200 described in S6 and S8.
CN201710889757.4A 2017-09-27 2017-09-27 Method for identifying radar one-dimensional range profile target based on convolution support vector machine Active CN107784320B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710889757.4A CN107784320B (en) 2017-09-27 2017-09-27 Method for identifying radar one-dimensional range profile target based on convolution support vector machine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710889757.4A CN107784320B (en) 2017-09-27 2017-09-27 Method for identifying radar one-dimensional range profile target based on convolution support vector machine

Publications (2)

Publication Number Publication Date
CN107784320A true CN107784320A (en) 2018-03-09
CN107784320B CN107784320B (en) 2019-12-06

Family

ID=61434087

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710889757.4A Active CN107784320B (en) 2017-09-27 2017-09-27 Method for identifying radar one-dimensional range profile target based on convolution support vector machine

Country Status (1)

Country Link
CN (1) CN107784320B (en)

Cited By (37)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108665000A (en) * 2018-05-03 2018-10-16 华东交通大学 A kind of digital picture automatic marking method based on analysis of uncertainty
CN108681698A (en) * 2018-04-28 2018-10-19 武汉大学 A kind of extensive iris identification method with privacy protection function
CN108776779A (en) * 2018-05-25 2018-11-09 西安电子科技大学 SAR Target Recognition of Sequential Images methods based on convolution loop network
CN108804581A (en) * 2018-05-24 2018-11-13 广州中大百迅信息技术有限公司 A kind of congener physical examination rope method and system based on deep learning
CN108957418A (en) * 2018-05-30 2018-12-07 西安电子科技大学 A kind of radar target identification method based on Recognition with Recurrent Neural Network model
CN108985268A (en) * 2018-08-16 2018-12-11 厦门大学 Conclusion formula high Resolution Range Profile Identification of Radar method based on depth migration study
CN108983187A (en) * 2018-07-11 2018-12-11 西安电子科技大学 Online radar target identification method based on EWC
CN108985454A (en) * 2018-06-28 2018-12-11 西南电子技术研究所（中国电子科技集团公司第十研究所） Airline carriers of passengers individual goal recognition methods
CN109709536A (en) * 2019-01-24 2019-05-03 电子科技大学 A kind of SAR moving target detection method based on convolutional neural networks
CN109753874A (en) * 2018-11-28 2019-05-14 南京航空航天大学 A kind of low slow small classification of radar targets method based on machine learning
CN110018453A (en) * 2019-03-28 2019-07-16 西南电子技术研究所（中国电子科技集团公司第十研究所） Intelligent type recognition methods based on aircraft track feature
CN110197218A (en) * 2019-05-24 2019-09-03 绍兴达道生涯教育信息咨询有限公司 Thunderstorm gale grade forecast classification method based on multi-source convolutional neural networks
CN110378385A (en) * 2019-06-20 2019-10-25 安徽省农业科学院畜牧兽医研究所 A kind of beef texture automatic measure grading method, system, device and storage medium
CN110687514A (en) * 2019-10-16 2020-01-14 电子科技大学 Nonlinear discrimination learning true and false target one-dimensional range profile feature extraction method
CN110764064A (en) * 2019-11-08 2020-02-07 哈尔滨工业大学 Radar interference signal identification method based on deep convolutional neural network integration
CN110826643A (en) * 2019-11-20 2020-02-21 上海无线电设备研究所 Offshore target identification method based on polarized Euler feature fusion deep learning
CN111062322A (en) * 2019-12-17 2020-04-24 西安电子科技大学 Phased array radar behavior recognition method based on Support Vector Machine (SVM)
CN111126455A (en) * 2019-12-05 2020-05-08 北京科技大学 Abrasive particle two-stage identification method based on Lightweight CNN and SVM
CN111368900A (en) * 2020-02-28 2020-07-03 桂林电子科技大学 Image target object identification method
CN111582236A (en) * 2020-05-27 2020-08-25 哈尔滨工程大学 LPI radar signal classification method based on dense convolutional neural network
CN111610517A (en) * 2020-06-09 2020-09-01 电子科技大学 Secondary radar signal processing method based on deep four-channel network
CN111814437A (en) * 2020-05-28 2020-10-23 杭州视氪科技有限公司 Method for converting braille into Chinese based on deep learning
CN111860585A (en) * 2020-06-11 2020-10-30 南京简凡科技有限公司 Automatic fatty liver detection method based on ultrasonic image artificial intelligence algorithm
CN111868803A (en) * 2018-03-23 2020-10-30 罗伯特·博世有限公司 Generating a composite radar signal
CN111898695A (en) * 2020-08-10 2020-11-06 武汉悟空游人工智能应用软件有限公司 Convenient convolution method
CN112214929A (en) * 2020-09-27 2021-01-12 电子科技大学 Radar interference suppression method for intermittent sampling repeated forwarding type interference
CN112364925A (en) * 2020-11-16 2021-02-12 哈尔滨市科佳通用机电股份有限公司 Deep learning-based rolling bearing oil shedding fault identification method
CN112966667A (en) * 2021-04-06 2021-06-15 中国人民解放军海军航空大学 Method for identifying one-dimensional distance image noise reduction convolution neural network of sea surface target
CN113298259A (en) * 2021-06-10 2021-08-24 中国电子科技集团公司第十四研究所 CNN network reasoning framework design method supporting embedded platform multi-core parallel
CN113537417A (en) * 2021-09-17 2021-10-22 南京隼眼电子科技有限公司 Target identification method and device based on radar, electronic equipment and storage medium
CN113960553A (en) * 2021-10-14 2022-01-21 电子科技大学 Gaussian weight distribution feature extraction method in one-dimensional image recognition
CN114070684A (en) * 2020-08-05 2022-02-18 西南交通大学 Modulation signal identification method based on two-dimensional convolutional neural network
CN114488140A (en) * 2022-01-24 2022-05-13 电子科技大学 Small sample radar one-dimensional image target identification method based on deep migration learning
CN114492505A (en) * 2021-12-24 2022-05-13 西安电子科技大学 Aerial group target and extended target identification method based on semi-measured data
CN114519372A (en) * 2022-01-28 2022-05-20 西安电子科技大学 One-dimensional range profile target identification method based on support vector machine
CN117111013A (en) * 2023-08-22 2023-11-24 南京慧尔视智能科技有限公司 Radar target tracking track starting method, device, equipment and medium
CN113298259B (en) * 2021-06-10 2024-04-26 中国电子科技集团公司第十四研究所 CNN (computer network) reasoning framework design method supporting multi-core parallelism of embedded platform

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016097758A1 (en) * 2014-12-19 2016-06-23 Apical Limited Sensor noise profile
CN106408030A (en) * 2016-09-28 2017-02-15 武汉大学 SAR image classification method based on middle lamella semantic attribute and convolution neural network
CN106599941A (en) * 2016-12-12 2017-04-26 西安电子科技大学 Method for identifying handwritten numbers based on convolutional neural network and support vector machine
CN107132516A (en) * 2017-05-16 2017-09-05 电子科技大学 A kind of Radar range profile's target identification method based on depth confidence network
CN107194433A (en) * 2017-06-14 2017-09-22 电子科技大学 A kind of Radar range profile's target identification method based on depth autoencoder network

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016097758A1 (en) * 2014-12-19 2016-06-23 Apical Limited Sensor noise profile
CN106408030A (en) * 2016-09-28 2017-02-15 武汉大学 SAR image classification method based on middle lamella semantic attribute and convolution neural network
CN106599941A (en) * 2016-12-12 2017-04-26 西安电子科技大学 Method for identifying handwritten numbers based on convolutional neural network and support vector machine
CN107132516A (en) * 2017-05-16 2017-09-05 电子科技大学 A kind of Radar range profile's target identification method based on depth confidence network
CN107194433A (en) * 2017-06-14 2017-09-22 电子科技大学 A kind of Radar range profile's target identification method based on depth autoencoder network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
MEI, SH 等: "FUSING DIFFERENT LEVELS OF DEEP FEATURES BY DEEP STACKED NEURAL NETWORK FOR HYPERSPECTRAL IMAGES", 《2017 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM (IGARSS)》 *
应厚泽: "基于机器学习的SAR图像分类与识别研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Cited By (53)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111868803A (en) * 2018-03-23 2020-10-30 罗伯特·博世有限公司 Generating a composite radar signal
CN108681698A (en) * 2018-04-28 2018-10-19 武汉大学 A kind of extensive iris identification method with privacy protection function
CN108681698B (en) * 2018-04-28 2021-03-16 武汉大学 Large-scale iris recognition method with privacy protection function
CN108665000A (en) * 2018-05-03 2018-10-16 华东交通大学 A kind of digital picture automatic marking method based on analysis of uncertainty
CN108804581A (en) * 2018-05-24 2018-11-13 广州中大百迅信息技术有限公司 A kind of congener physical examination rope method and system based on deep learning
CN108776779A (en) * 2018-05-25 2018-11-09 西安电子科技大学 SAR Target Recognition of Sequential Images methods based on convolution loop network
CN108957418A (en) * 2018-05-30 2018-12-07 西安电子科技大学 A kind of radar target identification method based on Recognition with Recurrent Neural Network model
CN108985454B (en) * 2018-06-28 2022-04-01 西南电子技术研究所（中国电子科技集团公司第十研究所） Individual target identification method for civil aviation airliner
CN108985454A (en) * 2018-06-28 2018-12-11 西南电子技术研究所（中国电子科技集团公司第十研究所） Airline carriers of passengers individual goal recognition methods
CN108983187A (en) * 2018-07-11 2018-12-11 西安电子科技大学 Online radar target identification method based on EWC
CN108985268B (en) * 2018-08-16 2021-10-29 厦门大学 Inductive radar high-resolution range profile identification method based on deep migration learning
CN108985268A (en) * 2018-08-16 2018-12-11 厦门大学 Conclusion formula high Resolution Range Profile Identification of Radar method based on depth migration study
CN109753874A (en) * 2018-11-28 2019-05-14 南京航空航天大学 A kind of low slow small classification of radar targets method based on machine learning
CN109709536A (en) * 2019-01-24 2019-05-03 电子科技大学 A kind of SAR moving target detection method based on convolutional neural networks
CN110018453B (en) * 2019-03-28 2023-05-23 西南电子技术研究所（中国电子科技集团公司第十研究所） Intelligent model identification method based on aircraft track characteristics
CN110018453A (en) * 2019-03-28 2019-07-16 西南电子技术研究所（中国电子科技集团公司第十研究所） Intelligent type recognition methods based on aircraft track feature
CN110197218B (en) * 2019-05-24 2021-02-12 绍兴达道生涯教育信息咨询有限公司 Thunderstorm strong wind grade prediction classification method based on multi-source convolution neural network
CN110197218A (en) * 2019-05-24 2019-09-03 绍兴达道生涯教育信息咨询有限公司 Thunderstorm gale grade forecast classification method based on multi-source convolutional neural networks
CN110378385A (en) * 2019-06-20 2019-10-25 安徽省农业科学院畜牧兽医研究所 A kind of beef texture automatic measure grading method, system, device and storage medium
CN110687514A (en) * 2019-10-16 2020-01-14 电子科技大学 Nonlinear discrimination learning true and false target one-dimensional range profile feature extraction method
CN110687514B (en) * 2019-10-16 2022-03-15 电子科技大学 Nonlinear discrimination learning true and false target one-dimensional range profile feature extraction method
CN110764064A (en) * 2019-11-08 2020-02-07 哈尔滨工业大学 Radar interference signal identification method based on deep convolutional neural network integration
CN110826643B (en) * 2019-11-20 2022-08-19 上海无线电设备研究所 Offshore target identification method based on polarized Euler feature fusion deep learning
CN110826643A (en) * 2019-11-20 2020-02-21 上海无线电设备研究所 Offshore target identification method based on polarized Euler feature fusion deep learning
CN111126455A (en) * 2019-12-05 2020-05-08 北京科技大学 Abrasive particle two-stage identification method based on Lightweight CNN and SVM
CN111126455B (en) * 2019-12-05 2023-07-14 北京科技大学 Abrasive particle two-stage identification method based on Lightvight CNN and SVM
CN111062322A (en) * 2019-12-17 2020-04-24 西安电子科技大学 Phased array radar behavior recognition method based on Support Vector Machine (SVM)
CN111368900A (en) * 2020-02-28 2020-07-03 桂林电子科技大学 Image target object identification method
CN111582236B (en) * 2020-05-27 2022-08-02 哈尔滨工程大学 LPI radar signal classification method based on dense convolutional neural network
CN111582236A (en) * 2020-05-27 2020-08-25 哈尔滨工程大学 LPI radar signal classification method based on dense convolutional neural network
CN111814437A (en) * 2020-05-28 2020-10-23 杭州视氪科技有限公司 Method for converting braille into Chinese based on deep learning
CN111610517A (en) * 2020-06-09 2020-09-01 电子科技大学 Secondary radar signal processing method based on deep four-channel network
CN111610517B (en) * 2020-06-09 2022-06-07 电子科技大学 Secondary radar signal processing method based on deep four-channel network
CN111860585A (en) * 2020-06-11 2020-10-30 南京简凡科技有限公司 Automatic fatty liver detection method based on ultrasonic image artificial intelligence algorithm
CN114070684B (en) * 2020-08-05 2024-02-09 西南交通大学 Modulation signal identification method based on two-dimensional convolutional neural network
CN114070684A (en) * 2020-08-05 2022-02-18 西南交通大学 Modulation signal identification method based on two-dimensional convolutional neural network
CN111898695A (en) * 2020-08-10 2020-11-06 武汉悟空游人工智能应用软件有限公司 Convenient convolution method
CN112214929A (en) * 2020-09-27 2021-01-12 电子科技大学 Radar interference suppression method for intermittent sampling repeated forwarding type interference
CN112214929B (en) * 2020-09-27 2022-12-02 电子科技大学 Radar interference suppression method for intermittent sampling repeated forwarding type interference
CN112364925A (en) * 2020-11-16 2021-02-12 哈尔滨市科佳通用机电股份有限公司 Deep learning-based rolling bearing oil shedding fault identification method
CN112364925B (en) * 2020-11-16 2021-06-04 哈尔滨市科佳通用机电股份有限公司 Deep learning-based rolling bearing oil shedding fault identification method
CN112966667A (en) * 2021-04-06 2021-06-15 中国人民解放军海军航空大学 Method for identifying one-dimensional distance image noise reduction convolution neural network of sea surface target
CN113298259A (en) * 2021-06-10 2021-08-24 中国电子科技集团公司第十四研究所 CNN network reasoning framework design method supporting embedded platform multi-core parallel
CN113298259B (en) * 2021-06-10 2024-04-26 中国电子科技集团公司第十四研究所 CNN (computer network) reasoning framework design method supporting multi-core parallelism of embedded platform
CN113537417A (en) * 2021-09-17 2021-10-22 南京隼眼电子科技有限公司 Target identification method and device based on radar, electronic equipment and storage medium
CN113960553A (en) * 2021-10-14 2022-01-21 电子科技大学 Gaussian weight distribution feature extraction method in one-dimensional image recognition
CN113960553B (en) * 2021-10-14 2023-04-18 电子科技大学 Gaussian weight distribution feature extraction method in one-dimensional image recognition
CN114492505A (en) * 2021-12-24 2022-05-13 西安电子科技大学 Aerial group target and extended target identification method based on semi-measured data
CN114488140B (en) * 2022-01-24 2023-04-25 电子科技大学 Small sample radar one-dimensional image target recognition method based on deep migration learning
CN114488140A (en) * 2022-01-24 2022-05-13 电子科技大学 Small sample radar one-dimensional image target identification method based on deep migration learning
CN114519372A (en) * 2022-01-28 2022-05-20 西安电子科技大学 One-dimensional range profile target identification method based on support vector machine
CN117111013A (en) * 2023-08-22 2023-11-24 南京慧尔视智能科技有限公司 Radar target tracking track starting method, device, equipment and medium
CN117111013B (en) * 2023-08-22 2024-04-30 南京慧尔视智能科技有限公司 Radar target tracking track starting method, device, equipment and medium

Also Published As

Publication number Publication date
CN107784320B (en) 2019-12-06

Similar Documents

Publication Publication Date Title
CN107784320A (en) Radar range profile's target identification method based on convolution SVMs
CN110414377B (en) Remote sensing image scene classification method based on scale attention network
CN104217214B (en) RGB D personage's Activity recognition methods based on configurable convolutional neural networks
CN105320965B (en) Sky based on depth convolutional neural networks composes united hyperspectral image classification method
CN108984745A (en) A kind of neural network file classification method merging more knowledge mappings
CN108647742A (en) Fast target detection method based on lightweight neural network
CN110516596A (en) Empty spectrum attention hyperspectral image classification method based on Octave convolution
WO2019179403A1 (en) Fraud transaction detection method based on sequence width depth learning
CN108681752A (en) A kind of image scene mask method based on deep learning
CN105279554B (en) The training method and device of deep neural network based on Hash coding layer
CN108280233A (en) A kind of VideoGIS data retrieval method based on deep learning
CN106991440B (en) Image classification method of convolutional neural network based on spatial pyramid
CN106991374A (en) Handwritten Digit Recognition method based on convolutional neural networks and random forest
CN107066973A (en) A kind of video content description method of utilization spatio-temporal attention model
CN108764460A (en) A kind of Time Series Forecasting Methods based on time convolution sum LSTM
CN114092832B (en) High-resolution remote sensing image classification method based on parallel hybrid convolutional network
CN107132516A (en) A kind of Radar range profile's target identification method based on depth confidence network
CN107330514A (en) A kind of Air Quality Forecast method based on integrated extreme learning machine
CN109523013A (en) A kind of air particle pollution level estimation method based on shallow-layer convolutional neural networks
CN109993164A (en) A kind of natural scene character recognition method based on RCRNN neural network
CN107798381A (en) A kind of image-recognizing method based on convolutional neural networks
CN106022355B (en) High spectrum image sky based on 3DCNN composes joint classification method
CN108446616A (en) Method for extracting roads based on full convolutional neural networks integrated study
CN109919202A (en) Disaggregated model training method and device
CN108920446A (en) A kind of processing method of Engineering document

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant